Hi Product Hunt — I built Holt because parallel coding agents create a blind spot before code reaches a pull request. Work can be spread across commits, the index, unstaged edits, untracked files, ignored paths and several Git worktrees. The risky question is not just “which branch is ahead?” It is “if this workspace is cleaned up now, does the project lose the last copy of anything?” Holt gives humans and agents one local decision surface for that question. It can: • show unique, redundant, conflicting and dependent work across worktrees; • inject timely project context through supported hooks and expose 16 decision-oriented MCP tools; • keep exact deletion authority separate from similarity and code-relationship heuristics; • preview cleanup, then use locked local quarantine rather than immediate deletion; • restore a quarantine with an exact command; • show the same evidence through a CLI, TUI and interactive graph. The free single-repository core is source-available under FSL-1.1-MIT for its defined Permitted Purposes, including internal commercial use that is not a Competing Use. I am most interested in rigorous feedback: what real multi-agent or multi-worktree failure should Holt handle that it does not handle today? If it produces either a false “safe” or an annoying false block, please send me the smallest reproduction.

Holt

Keep parallel coding-agent work visible and recoverable

Keep parallel coding-agent work visible and recoverable